OpenAI: GPT-5.4

OpenAI's GPT-5.4, launched on March 5, 2026, represents a significant advancement, unifying the Codex and GPT lines. It boasts an impressive 1M+ token context window (922K input, 128K output) supporting both text and image inputs, facilitating high-context reasoning and multimodal analysis. The model consistently ranks among the fastest, performing in the 88th percentile across eight benchmarks, and offers moderate pricing, placing it in the 25th percentile. Notably, GPT-5.4 demonstrates exceptional reliability with a 100% success rate across all benchmarks, indicating minimal technical failures. Performance across benchmarks highlights several strengths. It achieved perfect accuracy in Hallucinations, General Knowledge, and Ethics, often being the most accurate model at its price point and speed. Its Instruction Following and Reasoning capabilities are strong, scoring 83.0% and 90.0% respectively. While its Coding accuracy is 88.0%, it stands out as the most accurate among models of comparable speed. The model also performs well in Email Classification (99.0% accuracy) and Mathematics (89.0% accuracy). GPT-5.4 is designed as a robust solution for general-purpose tasks and software engineering, capable of generating production-quality code and executing complex workflows efficiently.